AITOP WeeklyIssue 61940 stories

GPT-6's Interactive UI, Haiku 5.5's Deep Price Cut, DeepSeek's $12B Round

Singapore·5 Oct – 11 Oct·Ten minutes

OpenAI pushed GPT-6 to ChatGPT's 1.2 billion weekly users, making replies render charts, buttons and forms so the interface itself becomes the model's output. Anthropic answered on cost: Claude Haiku 5.5 lands at $0.10/M input and $0.50/M output — roughly 90% below Haiku 4.5 — with a 1M-token context window and 128k max output. DeepSeek, meanwhile, is close to raising at least $12 billion at a possible $71 billion valuation, with V4.1 Flash scoring 72.9% on ARC-AGI-2 at $0.13 per task. For Singapore teams, the ceiling rose and the per-unit price fell in the same week, widening the selection window.

Five things this week

Enough to retell

GPT-6 向每周 12 亿 ChatGPT 用户全面推送，新增 Intelligent UI 能力

12 亿用户同时换到能生成界面的 GPT-6，交互范式变了

shao__meng

Anthropic 发布 Claude Haiku 5.5：价格降 90%，上下文扩到 100 万 token

Haiku 5.5 便宜九成、上下文 100 万，轻量场景可直接换

berryxia

OpenAI 发布 722 个 AI 生成的数学证明

722 个 AI 数学证明带 Lean 验证，矩阵乘法上界被压低

rohanpaul_ai

Google EmbeddingGemma 2 Model Released

EmbeddingGemma 2 enables local deployment for high-compliance sectors like healthcare, legal, finance, and government, addressing privacy concerns. For developers in Southeast Asia, this provides a cost-effective alternative to cloud-based multimodal RAG systems when paired with Gemma 4, reducing latency and data transfer costs for applications like local video search in regional languages.

AIGCLINK

OpenAI 发布内部前沿模型产出的多项新数学成果

OpenAI 公开 AI 生成的数学成果，还先咨询了 IAS 顾问组

OpenAI

US vs China

The same week, both sides

Dimension	United States	China	What it means for ASEAN
New models	GPT-6 rolled out to all ChatGPT users with Intelligent UI; Claude Haiku 5.5 and EmbeddingGemma 2 launched the same week.	DeepSeek V4.1 Flash hit 72.9% on ARC-AGI-2; StepFun's Step 5 Preview topped OpenRouter Trending.	ASEAN teams can price-compare US and Chinese models within the same day.
Prices	Claude Haiku 5.5 at $0.10/M input, $0.50/M output — about 90% cheaper than the prior generation.	DeepSeek V4.1 Flash costs $0.13 per ARC-AGI-2 task; V4 Flash usage spiked.	Falling inference prices make agent-style applications more economical.
Benchmarks	TasteVal uses 8 open tasks to quantify research taste; Opus 5.5 beat a 24-expert human baseline.	Tencent, Fudan and Tsinghua released ExplorationBench; task completion topped out at 73.4%.	Benchmarks are shifting to autonomous exploration; don't pick on leaderboards alone.
Policy	OpenAI consulted Princeton's IAS independent advisory group and cited its public recommendations before releasing math results.	Hackers used Chinese open-source agents to breach at least 7 South Korean financial firms, leaking session logs.	ASEAN financial firms need agent auditing and permission controls.